@minded-ai/mindedjs 2.0.6 → 2.0.7-beta-2

package/dist/browserTask/README.md

@@ -0,0 +1,419 @@
+# Browser Agent with CAPTCHA Bypass
+
+This implementation replaces the CLI-based browser-use approach with the Python SDK, providing enhanced capabilities including automatic CAPTCHA detection and bypass functionality.
+
+## 🚀 Features
+
+- **Python SDK Integration**: Uses browser-use Python SDK instead of CLI for better control
+- **CAPTCHA Bypass**: Automatic detection and solving of text-based CAPTCHAs
+- **Lifecycle Hooks**: Leverages browser-use lifecycle hooks for seamless operation
+- **Advanced OCR**: Uses OpenCV and Tesseract for robust CAPTCHA text extraction
+- **AI-Powered Solving**: Framework ready for LLM-based CAPTCHA solving (GPT-4V)
+- **Error Handling**: Comprehensive error handling and logging
+- **JSON Output**: Structured output format for better integration
+
+## 📁 File Structure
+
+```
+browserTask/
+├── browserAgent.py          # Main Python script with CAPTCHA bypass
+├── executeBrowserTask.ts    # Updated TypeScript wrapper
+├── requirements.txt         # Python dependencies
+├── setup.sh                # Setup script for dependencies
+└── README.md               # This documentation
+```
+
+## 🛠️ Setup
+
+### 1. Run the Setup Script
+
+The easiest way to get started is to run the setup script:
+
+```bash
+cd mindedjs/src/browserTask
+chmod +x setup.sh
+./setup.sh
+```
+
+This will:
+
+- Check Python 3.8+ installation
+- Install Python dependencies
+- Install Playwright browsers
+- Install Tesseract OCR
+- Make scripts executable
+
+### 2. Manual Setup (Alternative)
+
+If you prefer manual setup:
+
+```bash
+# Install uv (if not already installed)
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Create virtual environment
+uv venv --python 3.11
+
+# Activate virtual environment
+source .venv/bin/activate
+
+# Install Python dependencies
+uv pip install -r requirements.txt
+
+# Install Playwright browsers
+uv run playwright install
+
+# Install Tesseract OCR
+# macOS:
+brew install tesseract
+
+# Ubuntu/Debian:
+sudo apt-get install tesseract-ocr tesseract-ocr-eng
+
+# CentOS/RHEL:
+sudo yum install tesseract tesseract-langpack-eng
+```
+
+### 3. Configure API Keys
+
+Set your OpenAI API key (required for LLM functionality):
+
+```bash
+export OPENAI_API_KEY='your-api-key-here'
+```
+
+## 🎯 Usage
+
+### From TypeScript (MindedJS)
+
+The `executeBrowserTask` function now uses the Python implementation automatically:
+
+```typescript
+import { executeBrowserTask } from './browserTask/executeBrowserTask';
+
+const result = await executeBrowserTask('Navigate to example.com and fill out the contact form');
+console.log(result);
+```
+
+### Direct Python Usage
+
+You can also run the Python script directly:
+
+```bash
+# Activate virtual environment first
+source .venv/bin/activate
+
+# Basic usage
+python browserAgent.py -p "Navigate to google.com and search for AI"
+# Or with uv: uv run python browserAgent.py -p "Navigate to google.com and search for AI"
+
+# Form with CAPTCHA - the agent will automatically handle the CAPTCHA
+python browserAgent.py -p "Go to contact-form.com, fill out name as 'John Doe', email as 'john@example.com', message as 'Hello World', and submit the form"
+
+# Registration with CAPTCHA
+python browserAgent.py -p "Register on website.com with username 'testuser', email 'test@example.com', password 'securepass123', and complete the registration"
+
+# With custom max steps
+python browserAgent.py -p "Complete a multi-step process" --max-steps 50
+```
+
+**Note**: In the above examples, if the forms have CAPTCHAs, they will be automatically detected, solved, and filled. The agent will then continue with filling the other fields and submitting as instructed.
+
+## 🔒 CAPTCHA Bypass Capabilities
+
+### Integration with Browser-Use
+
+The CAPTCHA bypass is designed to work seamlessly with browser-use agents. When a CAPTCHA is detected:
+
+1. **CAPTCHA Detection**: Automatic detection on each page load
+2. **Silent Solving**: CAPTCHA is solved and filled in the background
+3. **Task Continuation**: Browser-use continues with your original prompt
+4. **No Interruption**: The user's task flow remains uninterrupted
+
+**Perfect for forms like:**
+
+- Registration forms with name + email + CAPTCHA
+- Contact forms with message + CAPTCHA
+- Login forms with credentials + CAPTCHA
+- Multi-step forms with CAPTCHA verification
+
+### Supported CAPTCHA Types
+
+1. **Text CAPTCHAs**: Alphanumeric text challenges
+2. **Simple Math**: Basic arithmetic problems
+3. **Image-based**: Framework ready for image selection CAPTCHAs
+
+### How It Works
+
+1. **Automatic Detection**: Lifecycle hooks monitor each page for CAPTCHA indicators
+2. **Image Processing**: OpenCV extracts and preprocesses CAPTCHA images
+3. **OCR Recognition**: Tesseract performs text recognition
+4. **Smart Input**: Automatically finds and fills CAPTCHA input fields
+5. **Continue Task**: Lets browser-use continue with original instructions (other fields, submission, etc.)
+
+### CAPTCHA Detection Indicators
+
+The system looks for these common CAPTCHA indicators:
+
+- `captcha`
+- `recaptcha`
+- `hcaptcha`
+- `verification`
+- `security check`
+- `prove you are human`
+
+## 🔧 Configuration
+
+### Environment Variables
+
+```bash
+# Required
+OPENAI_API_KEY=your-openai-api-key
+
+# Optional
+PYTHONPATH=/path/to/browserTask  # Set automatically by TypeScript wrapper
+```
+
+### Python Script Arguments
+
+```bash
+python3 browserAgent.py --help
+```
+
+- `-p, --prompt`: Task prompt (required)
+- `--max-steps`: Maximum number of steps (default: 30)
+- `--output-format`: Output format (text|json, default: text)
+
+## 🏗️ Architecture
+
+### Lifecycle Hooks Integration
+
+The implementation uses browser-use lifecycle hooks for seamless CAPTCHA handling:
+
+```python
+async def captcha_detection_hook(agent):
+    """Runs before each step to check for CAPTCHAs"""
+    page = await agent.browser_session.get_current_page()
+
+    # 1. Check for CAPTCHA indicators on the page
+    # 2. If found, solve and fill the CAPTCHA field
+    # 3. Let browser-use continue with the original task
+    # 4. No interruption to the user's workflow
+```
+
+**Flow Example:**
+
+```
+User Task: "Fill out the contact form with my details and submit"
+
+Step 1: Navigate to contact page
+Step 2: [Hook detects CAPTCHA] → Solve CAPTCHA → Fill CAPTCHA field
+Step 3: Fill name field (as per user instruction)
+Step 4: Fill email field (as per user instruction)
+Step 5: Fill message field (as per user instruction)
+Step 6: Click submit button (as per user instruction)
+```
+
+### CAPTCHA Bypass Flow
+
+```mermaid
+graph TD
+    A[Page Load] --> B[Lifecycle Hook Triggered]
+    B --> C{CAPTCHA Detected?}
+    C -->|No| D[Continue Normal Flow]
+    C -->|Yes| E[Take Screenshot]
+    E --> F[Extract CAPTCHA Region]
+    F --> G[OCR Text Recognition]
+    G --> H{Text Extracted?}
+    H -->|Yes| I[Fill CAPTCHA Field]
+    H -->|No| J[Try AI Solving]
+    J --> I
+    I --> K[Browser-use Continues Task]
+```
+
+### Tool Architecture
+
+```python
+class CaptchaBypass(Tool):
+    """Advanced CAPTCHA detection and solving tool"""
+
+    async def use(self, page, input_text="") -> ActionResult:
+        # Main entry point for CAPTCHA solving
+
+    async def _detect_and_solve_captcha(self, cv_img, pil_img):
+        # OpenCV + OCR processing
+
+    async def _ai_solve_captcha(self, screenshot):
+        # AI-powered solving (extensible)
+
+    async def _fill_captcha_solution(self, page, solution):
+        # Find inputs and fill solution (no submission)
+```
+
+## 🧪 Testing
+
+### Test Basic Functionality
+
+```bash
+# Activate virtual environment first
+source .venv/bin/activate
+
+# Test simple navigation
+python browserAgent.py -p "Go to google.com"
+
+# Test with potential CAPTCHA site
+python browserAgent.py -p "Navigate to a form and fill it out"
+```
+
+### Test CAPTCHA Bypass
+
+You can test the CAPTCHA bypass by:
+
+1. Finding a site with text CAPTCHAs
+2. Running the agent on that site
+3. Monitoring logs for CAPTCHA detection and solving
+
+### Debug Mode
+
+Enable detailed logging by modifying the Python script:
+
+```python
+logging.basicConfig(level=logging.DEBUG)  # Change from INFO to DEBUG
+```
+
+## 🔍 Troubleshooting
+
+### Common Issues
+
+1. **Python Dependencies**
+
+   ```bash
+   # With virtual environment activated
+   uv pip install --upgrade browser-use opencv-python pytesseract pillow
+   # Or: pip install --upgrade browser-use opencv-python pytesseract pillow
+   ```
+
+2. **Tesseract Not Found**
+
+   ```bash
+   # Verify installation
+   tesseract --version
+
+   # Add to PATH if needed
+   export PATH="/usr/local/bin:$PATH"
+   ```
+
+3. **Playwright Browser Issues**
+
+   ```bash
+   uv run playwright install --with-deps
+   # Or with activated venv: playwright install --with-deps
+   ```
+
+4. **Permission Issues**
+   ```bash
+   chmod +x browserAgent.py
+   chmod +x setup.sh
+   ```
+
+### Debugging CAPTCHA Issues
+
+1. Check `captcha_isolated.png` for extracted CAPTCHA images
+2. Review logs for OCR results
+3. Verify Tesseract language packs: `tesseract --list-langs`
+
+## 🚀 Extending the Implementation
+
+### Adding New CAPTCHA Types
+
+```python
+# Extend the CaptchaBypass class
+class AdvancedCaptchaBypass(CaptchaBypass):
+    async def _solve_image_captcha(self, page):
+        # Implement image selection logic
+        pass
+
+    async def _solve_audio_captcha(self, page):
+        # Implement audio CAPTCHA solving
+        pass
+```
+
+### Custom Lifecycle Hooks
+
+```python
+async def custom_hook(agent):
+    """Add custom logic to lifecycle hooks"""
+    page = await agent.browser_session.get_current_page()
+    # Your custom logic here
+
+# Use in agent
+agent.run(on_step_start=custom_hook)
+```
+
+### AI-Powered CAPTCHA Solving
+
+The framework is ready for AI-powered CAPTCHA solving. Implement the `_ai_solve_captcha` method:
+
+```python
+async def _ai_solve_captcha(self, screenshot: bytes) -> Dict[str, Any]:
+    # Convert screenshot to base64
+    screenshot_b64 = base64.b64encode(screenshot).decode('utf-8')
+
+    # Send to GPT-4V or similar vision model
+    response = await self.llm.ainvoke([
+        {"type": "text", "text": "Solve this CAPTCHA:"},
+        {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{screenshot_b64}"}}
+    ])
+
+    return {"success": True, "solution": response.content}
+```
+
+## 📝 Migration from CLI
+
+The new implementation maintains the same interface as the original CLI version:
+
+### Before (CLI)
+
+```typescript
+spawn('browser-use', ['-p', prompt]);
+```
+
+### After (Python SDK)
+
+```typescript
+spawn('python3', [pythonScriptPath, '-p', prompt, '--output-format', 'json']);
+```
+
+### Benefits of Migration
+
+1. **Better Control**: Direct access to browser-use APIs
+2. **CAPTCHA Bypass**: Automatic CAPTCHA handling
+3. **Lifecycle Hooks**: Fine-grained control over browser behavior
+4. **Error Handling**: Structured error reporting
+5. **Extensibility**: Easy to add new features
+6. **Performance**: Reduced overhead compared to CLI spawning
+
+## 📄 License
+
+This implementation follows the same license as the parent MindedJS project.
+
+## 🤝 Contributing
+
+When contributing to the browser agent:
+
+1. Test with multiple CAPTCHA types
+2. Ensure proper error handling
+3. Add logging for debugging
+4. Update documentation
+5. Follow the existing code style
+
+## 🆘 Support
+
+For issues and questions:
+
+1. Check the troubleshooting section
+2. Review logs for error details
+3. Test with the simplest possible case
+4. Verify all dependencies are installed
+
+The browser agent with CAPTCHA bypass provides a robust foundation for automated web interactions while handling common anti-bot measures.